Frequent Itemset Mining Algorithms for Knowledge Discovery-A Compendious Review of Various Approaches

Frequent Itemset Mining is a Data Mining task that has drawn the attention of researchers over the years. This concept is used in Market Basket Analysis in particular and Decision Support problems in general. In this paper, we have focused on the developments in this area so far. We start with an account of the algorithms that generate Candidate Itemsets. This class of algorithms proves costly, particularly in cases where there exist a large number of Itemsets. Then we describe the tree based Frequent Pattern algorithm that does not require Candidate Itemset Generation, thereby bringing the cost down. After that, we introduce lattice based algorithms in which fewer database scans are needed and hence I/O cost gets reduced. Then we discuss an algorithm that uses a single recursive function and simplifies its structure without worrying too much about the speed. Then we move on to have a look at an algorithm that uses hyperlinks and saves time and space. Then we describe computationally efficient algorithms for Closed Itemsets. Finally, we discuss some algorithms that leverage the inherent advantages of some special data representations to enhance efficiency. While keeping the nub and essence intact, we have avoided the original algorithmic and mathematical notations to keep it perspicuous, coherent, and comprehensible. However, a simplified block diagram has been given, wherever the nature of the algorithm permits it, to summarize the functioning of the algorithm briefly.